EP3446466B1 - Method for fast reconfiguration of gm clocks in the tsn network by means of an explicit teardown message - Google Patents

Description

Time-Sensitive Networking (TSN) is a set of standards that the Time-Sensitive Networking Task Group (IEEE 802.1) is working on.

The standards in the standardization define mechanisms for the transmission of data via Ethernet networks. The majority of the projects define extensions to the bridging standard IEEE 802.1Q. These extensions primarily address the transmission with very low transmission latency and high availability. Possible areas of application are convergent networks with real-time audio / video streams and, in particular, real-time control streams, which, for. B. in automobiles or in industrial plants for control.

Redundancy is the additional existence of functionally identical or comparable resources of a technical system, if these are normally not required for trouble-free operation. Resources can e.g. B. redundant information, motors, assemblies, complete devices, control lines and power reserves. As a rule, these additional resources serve to increase failure, functional and operational safety. Redundancy is therefore widespread in industrial communication, particularly in industrial communication networks that require a particularly short "switch-over time" in the event of an error, i.e. the time required in the event of a component failure to change (planned or unplanned) from a primary system to a standby system.

The TSN Task Group is currently working on the revision of the gPTP, i.e. the generalized Precision Time Protocol (IEEE 802.1 AS-2011). One of the main changes in the revised version is the addition of redundancy support to meet the requirements of real-time communication in industrial applications for high availability time synchronization.

Redundancy can be achieved in various ways in gPTP (generalized Precision Time Protocol, IEEE 802.1AS-2011) networks.

A PTP network consists of communicating clocks. Of these participating devices, the one that determines the most exact time is determined via the Best Master Clock (BMC) algorithm. This device serves as a reference clock and is called the Grandmaster Clock.

During operation, the master distributes the time signal to its slaves to determine the delay. For this purpose, a time stamp in the form of a sync message is sent from the master reference clock to the slave, which determines the time at which the time stamps are received at its own time. In addition, the slave repeatedly sends a delay request message to the master, the reception time of which is in turn sent back to the slave as a delay response message.
The master-to-slave delay and the slave-to-master delay are determined from the differences between the four time stamps. These values therefore contain the difference between the two clocks and the message runtime with opposite signs. The mean of the two values thus provides the directed offset to the master, which is ultimately used to synchronize the slave clock.
The compensation of the transit times is based on the assumption that the outward and return journey of synchronization messages have the same average transit times and only change slowly over time. The slave approaches the reference time of the master continuously using a control process. In particular, jumps back in the slave time are avoided.

One possibility is the so-called "Master Redundancy", which provides two active GrandMasters (GM). A synchronized time is published in parallel, so that if one of the two GM clocks fails, the network can still be synchronized by the other clock. This procedure requires the selection of two clocks as redundant GMs, with the "better" clock typically functioning as the primary GM and the second best clock acting as an immediately usable replacement (hot standby) GM.

Based on the Rapid Spanning Tree Protocol (RSTP), the BMCA determines a single GM clock by spanning a tree with the root of the best clock. However, using the existing BMCA to determine two GMs is considered error-prone and cannot be automated because two BMCA instances have to run on different gPTP domains and network management is required to preconfigure clock attributes to guarantee that the selected GMs in different domains are not at the same clock.

A recently submitted entry at an IEEE 802.1 standardization meeting on gPTP presents a new protocol, which will be called BMCA + in the following. BMCA + supports selection and maintenance of two BestMaster clocks on a single gPTP domain. It uses so-called peerto-peer Announce + messages which are the best master selection information for the best and the second best watch. BMCA + does not use a spanning tree algorithm like BMCA based on RSTP. BMCA + uses the sequence number (SN) and a hold time mechanism to avoid loops and detect errors. Dynamic reconfiguration in response to system changes is also supported.

Unlike the IS-IS (ystem I ntermediate S ystem to I ntermediate S Protocol) based method which obtains a kompletteen set of global information in the local database, the BMCA + algorithm supports the exchange and Storage of only selected results from the currently two best watches. The advantage is the reduced data traffic and improved scalability. However, the data exchange scheme of the BMCA + causes a problem with the stability in the GM reconfiguration if a current GM is downgraded to the subordinate clock due to changes in the clock attributes and a new GM clock has to be determined to replace the previous GM. In this case, the BMCA + has an information conflict, which takes an excessively long reconfiguration time, which is much longer than in other cases.

In the slide set " Support of Automatic Configuration in Working Clocks with Redundancy ", IEEE Draft Vol 802., No. V01, November 2015, authors Feng Chen, Franz-Josef Götz, Jürgen Schmitt, http://www.ieee802.org/1/files/ public / docs2015 / aschen-goetz-sync-redundancy-1115-v01.pdf the underlying setting is described with BMCA + and Announce + messages, as well as the implementation of a configuration (ie first initialization) of the GM in a communication network of the type mentioned.

The Figure 1 shows an example of a run of the BMCA + protocol on a network with four linearly arranged nodes for the selection of the best and the second best clock as redundant GMs. The arrow t indicates the passage of time. The horizontal arrows show the sending of messages between the network elements, including the information contained in the message. The GM new selection is triggered by the currently best clock on network node A, which lowers its priority value from 3 to 8. The nodes are in the Figure 1 and 2 labeled as follows: (A3: 4) means, for example, clock A with priority 3 and sequence number 4. A sends an announcement message (Announce +) with A8 as the second best GM clock. After receiving the change, each node must make a new selection locally and propagate these new selection results.

Unlike nodes A and B, which continue to view A8 as the second best GM watch, node C will send A8 with its own watch (C7: 4) in the current GM list and an announcement message without A8. This transaction provokes the next neighbor D, who knows nothing about the change in A. The result is that the old values of (A3: 4) stored in D are not deleted overwritten in a timely manner and therefore have the chance to overwrite the results of the new selection. This data inconsistency is noticeable as soon as A finds the old information in B's Announce + message. However, based on the current BMCA + protocol, A can only reply to B by resending the current results based on BMCA +. This does not resolve the conflict. As a result of the conflict between the obsolete and the new information on the same watch, the GM new selection cannot establish itself as long as the old information (A3: 4) is still available on the network. This deadlock is currently being intercepted by a timeout, but this takes longer than desired.

Like in the Figure 1 shown in the current solution, a timeout event later starts this cleaning action for each network node that has still stored the outdated information (A3: 4) and in the current example cause the network nodes to finally choose network node C as the new GM clock. This presupposes that the time lasts as long as the hold time, the value of which is usually several seconds. Selecting a much smaller value for the hold time, for example hundreds of milliseconds, can be helpful in reducing the reconfiguration time, but this is at the expense of significantly increased data traffic in the form of periodic Announce + messages that are transmitted.

It is an object of the invention to solve the problem described. This object is achieved by a method according to the features of independent claim 1 and a device claim according to claim 6.

The example from Figure 1 shows that the key to solving the problem is to perform a network wide clean up of the outdated information to ensure that outdated GM information with a higher priority cannot overwrite new information. The current BMCA + relies on the hold-time mechanism to quickly carry out the GM reconfiguration on all nodes.

The task is solved by introducing a concrete advertizing scheme for the BMCA + protocol in order to remedy the instability described during the GM reconfiguration period. The method uses an Announce + message for an extra notification, which shows a clock with outdated information and spreads the new information in the entire network. The goal is to inform all watches that outdated information is available so that the GM reconfiguration can be carried out quickly and without any additional conflicts. This so-called "teardown" notification is provided in the event that a current best clock (that is, either first or second best GM clock) downgrades its priority in order to become a subordinate clock. In other cases, if the best (or second best) watch increases its priority or a subordinate watch changes its attributes, it is not necessary to send a notification, since the problems described will not occur here.

• Type (Typ): bestimmt den Typ des Attributes
• Length (Länge): bestimmt die Übertragungslänge des Attributes
• Value (Wert): enthält den eigentlichen Wert des Attributes.

Figure 2 describes the advertising process based on the same network configuration as in Figure 1 shown. The downgraded master clock A must first send information (tear down) that contains the ClockID and the last serial number SN before the change. A new Announce + message is then sent, which contains the changed selection results (D5: 4, A8: 5) and the tear-down information (A: 4).
Instead of sending the full set of information in this message, e.g. B. (A3.4) the minimum of information is transmitted in the tear-down information, which for Selection of the redundant clock information is sufficient, i.e. ClockID and SN.
The tear-down message can be transmitted as TLV (type length value). The type length value format (TLV; German type length value) is used in network protocols and file formats to transmit a variable number of attributes in a message or file.
An attribute is transmitted by the following triple (type, length, value) at the end of the Announce + message:

• Type: determines the type of the attribute
• Length: determines the transmission length of the attribute
• Value: contains the actual value of the attribute.

When receiving an Announce + message with the tear-down message, each watch must first carry out a cleaning action in the local database and then select the best watch. Each watch must therefore forward the received tear-down message in the Announce + message to the nearest network node together with the updated selection data. The forwarding of a tear-down notification is unconditional, regardless of whether an intermediate watch detects that after the values have been updated, the watch that is downgraded continues to be accepted as one of the two best watches. For example, in Figure 2 the network node C a new selection result without A, but still passes the values (A3: 4) to the neighbor. This behavior ensures that the tear-down message is seen by all clocks in the network, so that the superfluous information can be completely removed from the network and all network elements.

The transmission of the BMCA + Announce + message uses an SN-based link-local information exchange mechanism to circulate the information about the communication of the best watch in the network, which is also used to advertise the tear-down notification. The port-based loop prevention action ensures that the same tear-down Message, identified by the pair "ClockID and SN", is only transmitted once via the same data port.

To further increase reliability, a watch that has received the new notification through one of its ports can acknowledge this receipt by sending an Announce + message to the original sender that contains the same Announce message.
With such a confirmation message, both network elements at the ends of the connection mutually confirm that they have been notified and now stop advertising the tear-down message in the following Announce + transmissions via the same link.
The confirmation messages in Announce + are in the Figure 2 not included.

As in Figure 2 shown, the Announce + message with the tear-down notification is propagated through the entire network. The superfluous best clock information is gradually deleted from each clock. Advantageously, the process described is duet faster than that with the prior art as in Figure 1 described, carried out. The time for GM reconfiguration is extremely reduced.

Figure 3 shows an alternative network structure to make it clear that in Figure 1 and 2 network shown is only to be understood as an example.

Claims (6)

1. Method for reconfiguring the information relating to redundant grand master clocks in a generalized Precision Time Protocol gPTP network by sending a teardown message from one network element (A, B, C, D) to an adjacent network element, which teardown message changes the information relating to at least one of the redundant grand master clocks (A, D) in the network elements of the gPTP network, wherein the teardown message is a first Announce+ message which, as information, contains only the clock ID and the serial number of the grand master clock before the change, wherein thereafter a new announcement message is sent, which is a second Announce+ message and contains changed results relating to a selection of a new best grand master clock and the information from the teardown message, and, after receiving the teardown message, the information relating to the affected grand master clocks is removed from the local database of the receiving network element and the selection of the new best grand master clock is carried out.
2. Method according to Patent Claim 1,
   wherein
   the teardown message is transmitted in the type-length-value TLV format.
3. Method according to one of the preceding patent claims,
   wherein
   the announcement message is transmitted only once via each data port of the network element.
4. Method according to one of the preceding patent claims, wherein
   the reception of the announcement message is acknowledged with a confirmation message to the transmitter.
5. Method according to Patent Claim 4,
   wherein
   the confirmation message contains a copy of the received message.
6. Network element which is incorporated in a communication network, wherein the communication network operates according to IEEE 802.1 TSN, wherein
   the network element is able and set up to operate according to one of the method claims.

Images